43590 AM
Background analysis
Otsu
Results (optimal threshold = 0.0510)
Sobel
ROC space
Sensitivity/Specificity against threshold
In this case (comparing with the other 2 images) we can say that the sensitivity and specificity change drastically if we vary the threshold from 0 to 10 and again from 20 to 60.
Roberts
ROC space
Sensitivity/Specificity against threshold
Canny
Here the behaviour of Canny is almost exactly the same as for 9343 AM and 10905 JL.
ROC space
See report for information on why Canny performed "badly". Even between sigma (0, 1] the sensitivity is under 0.40
Sensitivity/Specificity against sigma
Canny with Anisotropic Diffusion
For explanations please see the attached report.
Results (1, 15 and 20 iterations)
ROC space
CAD applied on 43590 AM performed worse than CAD applied on 9343 AM and 10905 JL. This happens when we have a low contrast image plus a non uniform illumination across the whole image (in this example border cells are very dark). Therefore border cells are removed when increasing the number of iterations which then results in lower sensitivity values.
Sensitivity/Specificity against number of iterations and kappa
Dilate-Erode method
For explanations please see the attached report.
ROC space
Sensitivity/Specificity against threshold
Here sensitivity and specificity change drastically for the reasons explained above (i.e. the nature of the given image).
Laplacian of Gaussian
For explanations please see the attached report.
ROC space
It can be noted that LoG did not perform very well with this image.
Sensitivity/Specificity against sigma
Difference of Gaussian
ROC space
Sensitivity/Specificity against minimum sigma
